The following sequence may cause the VM abort during migration:
1. RUN_STATE_RUNNING,MIGRATION_STATUS_ACTIVE
2. before call migration_completion(), we send migrate_cancel
QMP command, the state machine is changed to:
RUN_STATE_RUNNING,MIGRATION_STATUS_CANCELLING
3. call migration_completion(), and the state machine is
switch to: RUN_STATE_RUNNING,MIGRATION_STATUS_COMPLETED
4. call migration_iteration_finish(), because the migration
status is COMPLETED, so it will try to set the runstate
to POSTMIGRATE, but RUNNING-->POSTMIGRATE is an invalid
transition, so abort().
The migration_completion() should not change the migration state
to COMPLETED if it is already changed to CANCELLING.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Message-Id: <20201105091726.148-1-longpeng2@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Currently, the only difference of tcp channel and unix channel in
migration/socket.c is the way to build SocketAddress, but socket_parse()
can handle these two types, so use it to instead of tcp_build_address()
and unix_build_address().
The socket-type channel can be further unified based on the up, this
would be helpful for us to add other socket-type channels.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Message-Id: <20200806074030.174-2-longpeng2@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We found an issue when repeat reboot in guest during migration, it cause the
migration thread never be waken up again.
<main loop> |<migration_thread>
|
LOCK BQL |
... |
main_loop_should_exit |
pause_all_vcpus |
1. set all cpus ->stop=true |
and then kick |
2. return if all cpus is paused |
(by '->stopped == true'), else|
3. qemu_cond_wait [BQL UNLOCK] |
|LOCK BQL
|...
|do_vm_stop
| pause_all_vcpus
| (A)set all cpus ->stop=true
| and then kick
| (B)return if all cpus is paused
| (by '->stopped == true'), else
| (C)qemu_cond_wait [BQL UNLOCK]
4. be waken up and LOCK BQL | (D)be waken up BUT wait for BQL
5. goto 2. |
(BQL is still LOCKed) |
resume_all_vcpus |
1. set all cpus ->stop=false |
and ->stopped=false |
... |
BQL UNLOCK | (E)LOCK BQL
| (F)goto B. [but stopped is false now!]
|Finally, sleep at step 3 forever.
resume_all_vcpus should notice this race, so we need to move the change
of runstate before pause_all_vcpus in do_vm_stop() and ignore the resume
request if runstate is not running.
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Longpeng <longpeng2@huawei.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200316083732.2010-1-longpeng2@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
we found the following core in our environment:
0 0x00007fc6b06c2237 in raise ()
1 0x00007fc6b06c3928 in abort ()
2 0x00007fc6b06bb056 in __assert_fail_base ()
3 0x00007fc6b06bb102 in __assert_fail ()
4 0x0000000000702e36 in xhci_kick_ep (...)
5 0x000000000047897a in memory_region_write_accessor (...)
6 0x000000000047767f in access_with_adjusted_size (...)
7 0x000000000047944d in memory_region_dispatch_write (...)
(mr=mr@entry=0x7fc6a0138df0, addr=addr@entry=156, data=1648892416,
size=size@entry=4, attrs=attrs@entry=...)
8 0x000000000042df17 in address_space_write_continue (...)
10 0x000000000043084d in address_space_rw (...)
11 0x000000000047451b in kvm_cpu_exec (cpu=cpu@entry=0x1ab11b0)
12 0x000000000045dcf5 in qemu_kvm_cpu_thread_fn (arg=0x1ab11b0)
13 0x0000000000870631 in qemu_thread_start (args=args@entry=0x1acfb50)
14 0x00000000008959a7 in thread_entry_for_hotfix (pthread_cb=<optimized out>)
15 0x00007fc6b0a60dd5 in start_thread ()
16 0x00007fc6b078a59d in clone ()
(gdb) f 5
5 0x000000000047897a in memory_region_write_accessor (...)
529 mr->ops->write(mr->opaque, addr, tmp, size);
(gdb) p /x tmp
$9 = 0x62481a00 <-- last byte 0x00 is @epid
xhci_doorbell_write() already check the upper bound of @slotid an @epid,
it also need to check the lower bound.
Cc: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Longpeng <longpeng2@huawei.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 1556605301-44112-1-git-send-email-longpeng2@huawei.com
[ kraxel: fixed typo in subject line ]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Test-crypto-hash calls qcrypto_hash_bytesv/digest/base64 with
errp=NULL, this will cause a NULL pointer dereference if afalg_driver
doesn't support requested algos:
ret = qcrypto_hash_afalg_driver.hash_bytesv(alg, iov, niov,
result, resultlen,
errp);
if (ret == 0) {
return ret;
}
error_free(*errp); // <--- here
Because the error message is thrown away immediately, we should
just pass NULL to hash_bytesv(). There is also the same problem in
afalg-backend cipher & hmac, let's fix them together.
Reviewed-by: Eric Blake <eblake@redhat.com>
Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Longpeng <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This patch add a hmac speed benchmark, it helps us to
measure the performance by using "make check-speed" or
using "./tests/benchmark-crypto-hmac" directly.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This patch add a hash speed benchmark, it helps us to
measure the performance by using "make check-speed" or
using "./tests/benchmark-crypto-hash" directly.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Now we have two qcrypto backends, libiary-backend and afalg-backend,
but which one is faster? This patch add a cipher speed benchmark, it
helps us to measure the performance by using "make check-speed" or
using "./tests/benchmark-crypto-cipher" directly.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Adds afalg-backend hmac support: introduces some private APIs
firstly, and then intergrates them into qcrypto_hmac_afalg_driver.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Adds afalg-backend hash support: introduces some private APIs
firstly, and then intergrates them into qcrypto_hash_afalg_driver.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Adds afalg-backend cipher support: introduces some private APIs
firstly, and then intergrates them into qcrypto_cipher_afalg_driver.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The AF_ALG socket family is the userspace interface for linux
crypto API, this patch adds af_alg family support and some common
functions for af_alg backend. It'll be used by afalg-backend crypto
latter.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Maintainer: modified to report an error if AF_ALG is requested
but cannot be supported
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
1) makes the public APIs in hmac-nettle/gcrypt/glib static,
and rename them with "nettle/gcrypt/glib" prefix.
2) introduces hmac framework, including QCryptoHmacDriver
and new public APIs.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
1) Fix a handle-leak problem in qcrypto_hmac_new(), didn't free
ctx->handle if gcry_mac_setkey fails.
2) Extracts qcrypto_hmac_ctx_new() from qcrypto_hmac_new() for
gcrypt-backend impls.
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
1) makes the public APIs in hash-nettle/gcrypt/glib static,
and rename them with "nettle/gcrypt/glib" prefix.
2) introduces hash framework, including QCryptoHashDriver
and new public APIs.
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
1) makes the public APIs in cipher-nettle/gcrypt/builtin static,
and rename them with "nettle/gcrypt/builtin" prefix.
2) introduces cipher framework, including QCryptoCipherDriver
and new public APIs.
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Refactors the qcrypto_cipher_free(), splits it into two parts. One
is gcrypt/nettle__cipher_free_ctx() to free the special context.
This makes code more clear, what's more, it would be used by the
later patch.
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
ECB mode cipher doesn't need IV, if we setiv for it then qemu
crypto API would report "Expected IV size 0 not **", so we should
setiv only when the cipher algos really need.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
This patch add 3des-ede support for cryptodev. However this is effective
only when backend using libgcrypt/nettle, because cipher-builtin doesn't
support 3des-ede yet.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Single-DES is obsolete and it's broken/useless for decades, we should
remove it in cryptodev, as suggested by Daniel.
Guest who wants to use this obsolete cipher alg will use its built-in
implementation instead.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
As the key length of xts(aes) is different with other mode of aes,
so we should check specially in cryptodev_builtin_get_aes_algo, if
it is xts mode.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
This patch add nettle-backed HMAC algorithms support
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This patch add glib-backed HMAC algorithms support
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This patch add HMAC algorithms based on libgcrypt support
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This patch introduce HMAC algorithms framework.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This item will be used for support libcrypt-backed HMAC algorithms.
Support for hmac has been added in Libgcrypt 1.6.0, but we cannot
use pkg-config to get libcrypt's version. However we can make a
in configure to know whether current libcrypt support hmac.
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Libgcrypt and nettle support 3des-ede, so this patch add 3des-ede
support when using libgcrypt or nettle.
Reviewed-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Some software algorithms are based on the hardware's cache info, for example,
for x86 linux kernel, when cpu1 want to wakeup a task on cpu2, cpu1 will trigger
a resched IPI and told cpu2 to do the wakeup if they don't share low level
cache. Oppositely, cpu1 will access cpu2's runqueue directly if they share llc.
The relevant linux-kernel code as bellow:
static void ttwu_queue(struct task_struct *p, int cpu)
{
struct rq *rq = cpu_rq(cpu);
......
if (... && !cpus_share_cache(smp_processor_id(), cpu)) {
......
ttwu_queue_remote(p, cpu); /* will trigger RES IPI */
return;
}
......
ttwu_do_activate(rq, p, 0); /* access target's rq directly */
......
}
In real hardware, the cpus on the same socket share L3 cache, so one won't
trigger a resched IPIs when wakeup a task on others. But QEMU doesn't present a
virtual L3 cache info for VM, then the linux guest will trigger lots of RES IPIs
under some workloads even if the virtual cpus belongs to the same virtual socket.
For KVM, there will be lots of vmexit due to guest send IPIs.
The workload is a SAP HANA's testsuite, we run it one round(about 40 minuates)
and observe the (Suse11sp3)Guest's amounts of RES IPIs which triggering during
the period:
No-L3 With-L3(applied this patch)
cpu0: 363890 44582
cpu1: 373405 43109
cpu2: 340783 43797
cpu3: 333854 43409
cpu4: 327170 40038
cpu5: 325491 39922
cpu6: 319129 42391
cpu7: 306480 41035
cpu8: 161139 32188
cpu9: 164649 31024
cpu10: 149823 30398
cpu11: 149823 32455
cpu12: 164830 35143
cpu13: 172269 35805
cpu14: 179979 33898
cpu15: 194505 32754
avg: 268963.6 40129.8
The VM's topology is "1*socket 8*cores 2*threads".
After present virtual L3 cache info for VM, the amounts of RES IPIs in guest
reduce 85%.
For KVM, vcpus send IPIs will cause vmexit which is expensive, so it can cause
severe performance degradation. We had tested the overall system performance if
vcpus actually run on sparate physical socket. With L3 cache, the performance
improves 7.2%~33.1%(avg:15.7%).
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Longpeng(Mike)